RETURN_TO_BASE

Konkret
APP

ERP & AI System / Lead Architect / Full-Stack Engineer

ReactViteNode.jsSaldeo APISlack APIRAGVector DBLLM
≤15 min
Quote Prep Time
Real-time
Data Latency
3.2 GB
RAG Knowledge Base

> SYSTEM_OVERVIEW

Konkret APP is a dedicated web ERP system built for a general contractor in the construction industry. It replaced a fragmented ecosystem of spreadsheets, external project management tools and accounting software with a single intelligent cockpit — integrating real-time financial data from Saldeo, operational progress from task managers, and Slack alerting, all augmented by a RAG-powered AI quote engine trained on 70 historical cost estimates.

> SYSTEM_ARCHITECTURE

[Saldeo Smart API] ──(Webhooks / JSON)──┐
                                        │
[Aplikacja Tasków] ──(REST API / Sync)──┼─► [Konkret APP Core] ──► [Frontend (React/Vite)]
                                        │          │
[Slack API] ◄───────(Bots / Alerts)────┘          ▼
                                            [Silnik AI / RAG] ◄──► [Baza Wektorowa 3.2 GB]

> APP_SCREENSHOTS

Konkret APP – main dashboard with construction projects and budget KPIs
Dashboard – Twoje Budowy
Budget analysis module with Expenses vs Budget charts
Analiza Budżetowa
AI Wyceny RAG module – intelligent quote assistant
AI Wyceny (RAG Engine)

Core Modules

Central Dashboard & Real-Time Integrations

Aggregates key financial KPIs across all construction projects in real time. Auto-syncs invoices from Saldeo Smart API, categorises line items by construction phase (Foundations, Reinforcement, Walls) and maps every cost to its corresponding milestone — no manual data entry.

Budget Analysis Module

Comparative bar charts (Expenses vs Budget) and a cumulative Plan Utilisation curve for all active projects in parallel. Tabular breakdowns with automatic status flagging: green 'W normie' badge when costs stay within margin; instant red alert when a construction phase exceeds 90% of its budget.

AI Quote Engine (RAG)

A conversational AI assistant backed by a 3.2 GB vector knowledge base of 70 complete historical cost estimates and material catalogues. Answers natural-language Polish queries ('Prepare a shell estimate for a 150 m² semi-detached house') with contextually grounded drafts — no hallucinations, only real historical data.

Smart Notification Ecosystem

Event-driven architecture connects financial events to the team's Slack workspace. Every status change on an invoice (e.g. Cegła-Max 45 000 zł → AWAITING) or budget threshold breach triggers a formatted alert pushed to the project-specific Slack channel within seconds.

Business Results

AreaBeforeAfter (Konkret APP)
Quote preparation time2–4 business days (manual work)A few minutes (AI draft + review)
Cost data latencyUp to 2 weeks (waiting for accounting)Real-time (auto-sync with Saldeo)
Phase margin controlReactive (after the fact)Proactive (instant budget breach alerts)
Access to historical knowledgeScattered PDFs & spreadsheets on drivesCentralised RAG base accessible via chat

/// ENGINEERING_CHALLENGES_LOG

ERR_01

Saldeo API & Automatic Cost Categorisation

Invoices arrive as raw JSON webhooks from Saldeo. A custom categorisation algorithm parses line-item descriptions (e.g. 'Beton towarowy B25', 'Pręty żebrowane fi 12') using keyword matching and fuzzy logic to map each spend to the correct construction phase defined in the database — eliminating hours of manual book-keeping.

ERR_02

RAG Pipeline & Hybrid Search

70 PDF/XLSX cost estimates (3.2 GB) were structurally parsed — not treated as raw text. Custom chunking preserved the 'Position → Material/Labour → Quantity → Unit Price' table structure with metadata tags (year, building type, location). Hybrid dense + sparse retrieval ensures both semantic relevance and exact material-name precision.

ERR_03

Inflation-Adjusted Post-Processing

LLMs struggle with live price updates. The orchestration layer separates AI reasoning (context retrieval + structure generation) from mathematics. After the RAG draft is produced, the backend applies a dynamic percentage scalar defined by the user in the UI, recalculating all historical prices to current-quarter market rates with guaranteed numeric accuracy.

/// SERVICES_USED

Services behind the build

The engineering capabilities applied in this case study are available as standalone services.

Web Engineering

Digital brutalism architecture. Sites that are not templates, but manifestos.

View service

AI & Automation

Virtual employees who never sleep. Autonomous agents and workflows.

View service
/// MORE_WORK

Other case studies

SaaS / AI

AnswerLyzer AI Analytics

'LLM-as-a-Judge' architecture for automated brand visibility and sentiment scoring across ChatGPT, Gemini and Perplexity.

View case studyFintech / AI

PozyczkoBank Fintech AI

Smart SPA Fallback architecture combined with Python Serverless and 90% editorial automation via Gemini AI.

View case studyAI / E-Commerce

GiftFinder AI Platform

Autonomous agentic system generating 11,000+ SEO landing pages with hybrid Gemini + GPT-4o architecture.

View case study
/// KNOWLEDGE_BASE

Related reading

AI & Automation

How AI Reads Invoices from Email and Enters Them into ERP

AI can automatically read an invoice from an email attachment — PDF, scan, or phone photo — and enter the data directly into an ERP system without any manual retyping. Full automation of cost invoice processing: from the mailbox to accounting.

10 minAI & Automation

How to Build a Company Internal Knowledge Base with AI (RAG in Practice)

An internal knowledge base built on RAG lets you create your own company chatbot that answers only from your company's documents — not the model's guesses. Safe, up-to-date, precise AI with full control over your data.

11 min

Terminate
Silence

Initiate protocol. Establish connection. Let's build something loud.

> WAITING_FOR_INPUT...
WARSAW, PL
pawel.wiszniewski95@gmail.com